Machine learning in trading: theory, models, practice and algo-trading - page 1641
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
The problem looks not quite formalized - the set of parameters is not clear. Is the complete set of systems finite, countable, or continuous? Is the portfolio of fixed size? Is the system included in the portfolio with some weights or just yes/no?
Hmmm... Honestly, thanks for the question - as they say half the answer is already there
set of systems is countable and finite, there are no weights, and it is not planned - all are equal,
no miracle happened, the main problem arising in a simple TS is drawdown, the goal is not to minimize the drawdown by adding another TS - it does not matter at the time of drawdown will work 2 TS, or hope that there will be an alternative TS to replace the drawdown TS - this is an increase in risk, not looking there, I have already looked
the aim - to run the TS from the portfolio, but after the virtual testing AND after drawdown - there is a sense, according to the tests of TS - drawdowns are periodic and there is some time after drawdown, when the TS works - then the problem of how long to let this TS work, in this task is likely not the GA assistant, we need a certain intellectual component
Imho of course, but in my opinion it is necessary to dance from the "physics" of quotations of financial instruments. Their main property, in my opinion, is the change, sometimes very fast and cardinal, of statistical characteristics of a time series. In this sense it would be reasonable first of all to create a classifier that would sort the history into sections with similar statistical characteristics and give them numbers from 1 to 20, say. And then for each such similar type of market to create its own individual TS. But how to think up predictors for such partitioning of time series into sections with similar statistical characteristics - I can't really imagine.
wise idea...
I have turned to the zigzag... But since the beginning of March, they are simply not comparable to what they were before March. If before it could take half an hour or an hour to plot the knee, it now takes only 5 minutes to plot it with the same parameters due to the high volatility. So it makes no sense to study on the data before March. Everything is different now.
We should still think of something universal for high and low volatility.
Maybe something wave like that. The waves have remained but they have become wider.
It does not make sense to work with fixed parameters, not to learn until March!!!
Imho of course, but in my opinion we should rely on the "physics" of financial instruments quotes. Their basic property, to my mind, is the change, sometimes very fast and cardinal, of the statistical characteristics of a time series. In this sense it would be reasonable first of all to create a classifier that would sort the history into sections with similar statistical characteristics and give them numbers from 1 to 20, say. And then for each such similar type of market to create its own individual TS. But how to think up predictors for such partitioning of time series into sections with similar statistical characteristics - I can't really imagine.
Usually for this purpose the dynamics of change of MO of a series is used.
If ME changes insignificantly - "flat".
If ME is growing at a rate higher than X - "growing trend".
If ME is decreasing at a rate higher than X - "decreasing trend".
I also saw a classification based on dispersion.
It makes no sense to work with fixed parameters , not to train to March!!!
An example of a suitable indicator for MO without fixed parameters is possible?
https://www.youtube.com/watch?v=TykEeAM6v9U
https://www.youtube.com/watch?v=2JgoeuM7iVMThe set of systems is countable and finite, there are no weights, and it is not planned - all are equal,
the miracle did not happen, the main problem arising in a simple TS is a drawdown, the goal is not to minimize the drawdown by adding another TS - no matter if 2 TS at the time of drawdown, or hope that there will be an alternative TS to replace the drawdown TS - this is an increase in risk, not looking there, have looked
the goal - to run the TS from the portfolio, but after the virtual testing AND after drawdown - there is a sense, according to the tests of TS - drawdowns are periodic and there is some time after drawdown, when the TS works - then the problem of how long to let this TS work, in this task is likely not the GA assistant, you need a certain intellectual component
Perhaps, as you wrote earlier, you can do with standard means of MT tester. To be honest, I do not see in neural networks, in themselves, anything particularly remarkable. I suppose it's not worth avoiding the possibility to do without them.)
https://www.youtube.com/watch?v=TykEeAM6v9U
https://www.youtube.com/watch?v=2JgoeuM7iVMWe only have the quotes. Assuming it is a noisy signal, what is the reference signal?
For these examples you need an exemplary unnoised signal to calculate the correlation coefficients to clean up the noisy signal.
We only have quotes. Assuming it is a noisy signal, what is the reference signal?
This is your target function (just like in AMO), what you want to get as a result of filtering, want to remove noise ? describe your ideal signal and feed it as a reference , want to describe a trend ? same thing.
want to know the "ideal zigzag parameters" at the moment ? describe what the "ideal zigzag parameters" are for you then
try to get "IPZ" on each candle, I think you'll be interested to see it:)
And then you can even try to predict the half-scientific series of "ZPI" by the same ill-fated AMO ))
As a result you get an adaptive system with adequate zigzag parameters and one step ahead forecast, this is what poorIgor Makanu has been looking for a year and still can't find)) at least he's got it written in front of his nose. Also dearIgor Makanu this is a solution to your problem "when the system stops working", you can simply track the error AMO (based on the parameters zzz etc.) in real time, this will be your criterion of the efficiency of the system